ii — the labors

Workflows are data, not code.

Every automation here is a set of YAML definitions that compile into a typed graph before anything runs. They diff in review like any other file, and a mismatched pair of tasks is caught when you plan, not at three in the morning when it runs. The engine never grows a branch for your business, so your hundredth automation costs what your first did.

pipelines/audit.yamlwhat you write

kind: triggerversion: 1source: webhook---kind: taskversion: 1runner:  kind: agent  spec: audit.analyst@3input: report.request@1output: report.spec@1uses:policy:  timeout: 90s  retry: 2  idempotent: true---kind: taskversion: 1runner:  kind: script  runtime: python  entry: render.maininput: report.spec@1

ptn applywhat gets built

report.request@1trigger · webhook

derived from on:

gather@1task · agent

derived from uses:

ledgerpostgres · query

receiptsobject store · get

fx_rateshttp · request

derived from then:

render@1task · script

declared output

report.htmlvalue · file

You never drew an edge. Every arrow above came out of an on:, then:, or uses: reference — hover either side to see which.

four primitives — the whole vocabulary

Trigger: cron · webhook · manual
Task: script · agent
Resource: postgres · object store · http
Approval: approvers · timeout

five values — everything that crosses a seam

Text: a body of words
File: content-addressed blob handle
Table: columns plus a row source
Record: typed against a registered schema
Error: typed failure, routed not thrown

That is the entire vocabulary, and it is closed on purpose. All your business meaning lives in Records, checked against schemas you register, so the engine never grows a branch for your domain and your hundredth workflow type costs the same as your first.